Determinants of increasing COVID-19 fatality rates after SARS-CoV-2 P.1 emergence in Brazil
Andrea Brizzi, Ricardo P Schnekenberg, Iwona Hawryluk, Luciana Servo, Rafael H. M. Pereira, Carlos Prete, Marcia Castro, Charlie Whittaker, Melodie Monod, William M. Souza, Raphael Sonnabend, Alexandra Blenkinsop, Xiaoyue Xi, Lewis F. Buss, Swapnil Mishra, Bruce Nelson, Thomas Mellan, Seth Flaxman, Renan P Souza, Renato S. Aguiar, Ester C. Sabino, Samir Bhatt, Nuno R. Faria, Oliver Ratmann
Session info
This file presents the results of Stan model
## [1] "age_hfr_210619b.stan"with job tag
## [1] "hfr-fit-210619b1-resources-ex0-trvaluelogsc-hd1-htprivate"Objectives
Several studies have reported significantly higher, crude in-hospital fatality rates following COVID-19 attributable hospital admission across Brazil after P.1 detection. The overall, primary objective of this study was to
- assess infection severity of SARS-CoV-2 P.1 in the context of increased hospital admissions and increasing health care resources across most of Brazil
More detailed objectives were to
- characterise time trends in COVID-19 attributable fatal outcomes
- identify time trends in in-hospital fatality rates before and after P.1 emergence
- characterise time trends health care resources
- characterise differences in the absolute magnitude in in-hospital fatality rates across Brazil based on health care resources
- determine the relative contribution of P.1 and increasing health care demand due to a more infectious P.1 to increases in crude in-hospital fatality rates
Methods
We
- Performed a longitudinal observational study across Brazilian state capitals, using hospital admissions of severe acute respiratory infections and associated clinical outcomes reported to the SIVEP-Gripe platform. Data on population denominators and health care resources are from the Instituto de Pesquisa Econômica Aplicada, Ipea, and several databases from the Ministry of Health.
- Sequence data from GISAID was used to identify state capitals in which P.1 was detected by March 31 2021. P.1 was detected in 14 state capitals. Analyses were performed independently for each of the 14 cities to guard against the potential presence and impact of unmeasured confounders to this observational analysis.
- To disentangle the impact of P.1 versus increasing health care demand, we performed a modelling study. Briefly:
- We used GISAID data to calculate the frequency of P.1 infections over time in each city, and developed a multi-strain model to estimate local replacement dynamics. This allowed us to decompose the COVID-19 attributable deaths and COVID-19 attributable hospital admissions by SARS-CoV-2 variant, and then to calculate fatality rates per variant.
- We linked in the model fatality rates to health care demand predictors that describe demand per unit resource. This allowed us to estimate associations between fatality rates, health care demand, and health care resources, and to determine the relative contribution of P.1 versus demand per resources to fatality rates.
Results
Population denominators
As population denominators, we used 2020 population size projects from the PNADc survey of the Instituto Brasileiro de Geografia e EstatÃstica (IBGE). Population denominators were not consistent with individual-level data on vaccine administrations in each location, which could be due to unreported location of residence at time of vaccine administration. We adjusted population denominators upwards so that at most 99 percent of the population had received a first dose, and never adjusted population denominators downwards.
COVID-19 attributable fatal outcomes
Reported COVID-19 attributable deaths were stratified by location of death (out of hospital, private hospitals, public hospitals, private or public hospitals). Reported deaths in hospitals were adjusted upwards to account for patients with unkwnown outcomes. Censoring-adjusted deaths were compared to excess deaths:
Age composition of COVID-19 attributable fatal outcomes
We find substantial changes in the age composition of COVID-19 attributable deaths over time, since vaccine roll-out and since P.1 emergence:
Cities 1
Cities 2
Cities 3
Vaccine roll-out
In our modelling we adjusted for protection from fatal outcomes 2 weeks after vaccine administration:
Cities 1
Cities 2
Cities 3
Health care demand predictors
ICU beds per 100,000
Physicians per 100,000
Specialist physicians per 100,000 (cardiologists, anesthesiologists, intensive care)
Ventilators per 100,000
Proportion of residents among hospital admissions in this and past four weeks
SARI admissions in this and next four weeks per 100,000
SARI admissions in this and next two weeks per ICU bed
SARI admissions in this and next two weeks per physician
ICU admissions in this and next two weeks per ICU bed
ICU admissions in this and next two weeks per physician
ICU admissions in this and next four weeks per specialist physician
ICU admissions in this and next two weeks per ventilator
Out of hospital deaths in this and next four weeks
P.1 replacement dynamics by city
P.1 genotype frequency
COVID-19 attributable deaths
Time trends in in-hospital fatality rates by city (only age group 40-49)
Time trends in in-hospital fatality rates by city (all ages)
Belo Horizonte
Curitiba
Goiania
Manaus
Natal
Rio de Janeiro
Salvador
Sao Paulo
Association in-hospital fatality rates <-> health care demand predictors
Location effect to in-hospital fatality rates
Ratio in-hospital fatality rates in reference week across locations
In-hospital fatality rates in reference week versus health care demand predictors
Multiplier to in-hospital fatality rates based on health care demand predictors
Scenario compared to reference week, multiplier to in-hospital fatality rates
Scenario compared to reference week, deaths averted
Scenario compared to best city, multiplier to in-hospital fatality rates
Scenario compared to best city, excess deaths
P.1 effect to in-hospital fatality rates
Absolute in-hospital fatality rates by variant
Ratio in-hospital fatality rates P.1 vs non-P.1
Ratio share of age groups P.1 vs non-P.1
Model fits by city
Belo Horizonte
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Curitiba
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Goiania
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Manaus
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Natal
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Rio de Janeiro
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Salvador
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
Sao Paulo
Fit to genotype data
Fit to hospital admission data
Fit to death data
Fit to share of age groups in deaths
In-hospital fatality rate multipliers by city
Belo Horizonte
Curitiba
Goiania
Manaus
Natal
Rio de Janeiro
Salvador
Sao Paulo
